Computer-based rhythm diagnosis and its possible influence on non-expert ECG readers

Background: Systems providing computer-based analysis of the resting ECG seek to improve the quality of health care by providing accurate and timely automatic diagnosis of e g cardiac rhytm to clinicians. The accuracy of these diagnoses, however, remains questionable. Objectives: We tested the hypothesis that (a) two independent automated ECG systems have better accuracy in rhythm diagnosis than non-expert clinicians and (b) both systems provide correct diagnostic suggestions in a large percentage of cases when the diagnosis of non-expert clinicians is incorrect. Methods: 500 ECGs were manually analyzed by 2 senior experts, 3 non-expert clinicians and automatically by 2 automated systems. The accuracy of the non-expert rhythm statements was compared with the accuracy of each system statement. The proportion of rhythm statements when the clinician’s diagnoses were incorrect and the systems instead provided correct diagnosis was assessed. Results: 420 sinus rhythms and 156 rhythm disturbances were recognized by expert reading. Significance of the difference in accuracy between non experts and systems was p= 0.45 for system A and p= 0.11 for system B. The percentage of correct automated diagnoses in cases when the clinician was incorrect was 28 ± 10 % for system A and 25 ± 11% for system B (p = 0.09). Conclusion: The rhythm diagnoses of automated systems did not reach better average accuracy than those of non-expert readings. The computer diagnosis of rhythm can be incorrect both in cases where the clinicians fails in reaching the correct ECG diagnosis of heart rhythm.